This thesis describes the theoretical and computational background of the work carried out on the analysis of Single Particle Tracking (SPT) trajectories. The thesis is organized into five chapters. The first chapter introduces SPT, the experimental and computational pipeline used to reconstruct trajectories from microscopy videos, and the state of the art, with particular attention to how machine learning can be used to analyze trajectories. The second chapter gives a mathematical description of Brownian motion and anomalous diffusion. The third chapter introduces convolutional neural networks and Transformers, the two architectural families most relevant to the work. The fourth chapter explains conformal prediction as a tool for assigning uncertainty intervals to model outputs.